Debt Priority Structure and Bank Earnings Opacity

Transcription

1 Debt Priority Structure and Bank Earnings Opacity Abstract We examine how changes in debt priority structure that expose junior creditors to greater risk in the event of bankruptcy and thereby give them a greater monitoring incentive, affect banks earnings opacity. Our tests exploit a natural experiment in U.S banking where statelevel depositor preference laws provide exogenous variation in the priority of debt claims for creditors of state- but not nationally-chartered banks. Using difference-in-difference estimations and several measures of opacity, we document that subordinating claims of general creditors triggers economically and statistically significant reductions in earnings opacity. Consistent with the idea that junior claimants have greater monitoring incentives, we show that general creditors demand higher risk premiums which motivates banks to become less opaque to limit the increase in the costs of subordinated funds. Our findings illustrate the role of debt seniority in regulatory frameworks. Keywords: earnings opacity; monitoring; debt priority structure; natural experiment JEL Codes: G21, G28 1

2 1. Introduction The opacity of bank balance sheets impedes market discipline because it limits outsiders ability to accurately value banks and assess bank risk. In particular, information asymmetries arising from opacity have potential to undermine confidence in individual banks ability to raise capital, lead to the drying up of interbank markets, and ultimately fuel contagion, and increase systemic risk. For those reasons, policy makers are keen to design regulation that reduces such uncertainty and mitigates bank opacity. Typically, these policies require increased disclosure and come in the form of restrictions on asset composition and complexity. A complementary way to reduce bank opacity, in particular earnings opacity, is to limit banks scope for discretionary accounting choices that allow banks to delay the recognition of expected losses. Providing accurate incentives for debtholders to monitor bank conduct more closely is an avenue that has potential to achieve this objective. In this paper, we present the first test of whether changing the priority structure of debt claims may affect the degree of banks earnings opacity, measured by discretionary loan loss provisions (LLP) as described by Beatty and Liao (2014). Specifically, we exploit plausibly exogenous shocks that arise from the staggered nature of the enactment of depositor preference laws that change the priority of debt claims on the assets of failed banks in 15 U.S. states between 1983 and Depositor preference laws disproportionally affect different classes of creditors: while insured depositors position in the claim structure remains unaffected, the position of uninsured depositors and nondepositors changes because claims of uninsured depositors become senior to those of non-depositors. In contrast, depositor preference laws make non-depositors claims more junior, so that they are exposed to greater losses in the event of bankruptcy. Since debt priority structure has implications for debtholders monitoring incentives, any expected increase in monitoring by non-depositors following the subordination of their claims is likely to limit banks scope to engage in discretionary accounting choices and therefore reduce earnings opacity. Our experimental setup is econometrically appealing because depositor preference laws only affect state-charted banks but do not apply to nationally-chartered banks. We exploit the panel structure of the data to compute the average treatment effect based on comparisons of the state-chartered (i.e., the treatment) and nationally-chartered banks (i.e., the control group) within the same state-quarter, i.e. within identical macroeconomic environments. A large literature documents close relationships between banks earnings opacity, accounting rules for loan loss provisions, and bank risk. Early work by Beatty et al. (1995) shows that loan charge-offs and provisions are reflective of banks capital management and subsequent research by Collins et al. (1995) examines how cross-sectional bank characteristics such as capital and earnings correlate with variations in loss provisions. They argue that declines in loan loss provisions in years of low nondiscretionary earnings are consistent with earnings management. In a recent review, Bushman (2014) summarizes that discretionary choices in accounting adversely affect bank stability via accounting numbers that play a key role for the calculation of regulatory numbers and also by dampening market discipline. More specifically, Bushman and Williams (2012, 2015) highlight that delayed expected loan loss recognition and accounting discretion affect bank risk, and this association tends to be more pronounced during crises (Cohen et al. (2014)). Likewise, Huizinga and Laeven (2012) show that U.S. banks became increasingly opaque because they understated loan loss provisions and other write-downs during the recent crisis, and Iannotta and Kwan (2014) show that earnings opacity increases as a result of delayed loan loss recognition. 2

3 Ultimately, opacity limits outside investors ability to monitor banks (Acharya and Ryan (2016)). Further undesirable consequences that arise from information asymmetries originating from earnings opacity have been documented for non-financial firms by Bhattacharya et al. (2003). Aboody et al. (2005) show that earnings opacity increases the cost of equity, and Pittman and Fortin (2004) find that monitoring from Big-6 auditors decreases the cost of debt financing. Our work contributes to this literature by examining whether monitoring from bank creditors with different monitoring incentives plays any role for banks earnings opacity. This is a timely and policy-relevant question because Beatty and Liao (2014) argue that government bailouts, blanket guarantees, and other forms of implicit and explicit support during the recent crisis have weakened the monitoring incentives of depositors. This research is important for the following four reasons. First, the recent financial crisis has reinvigorated interest in the question of whether banks are opaque, and, if so what the consequences are which arise from increased earnings opacity. Examining ratings disagreements for bank bonds, Morgan (2002) argues that banks are indeed more opaque than their non-financial counterparts. In contrast, studies by Flannery et al. (2004, 2013) which home in on market microstructure properties of bank share prices do not fully support this conclusion. Gorton (2013) and Jayaraman and Thakor (2015) revisit the question of whether the nature of bank s business activities makes them optimally opaque. Gorton (2013) points out that opacity of bank securities in the US started in the 19 th century as a result of the need to prevent bank runs. His argument is based on the observation that bank coalitions, central banks, and the government have attempted to suppress bank-specific information by shutting down markets for bank liabilities during financial crises. Therefore, banks are indeed opaque and, while bank-specific information should be produced, such information should be filtered by regulators to reduce the risk of bank runs. Likewise, recent theory by Jayaraman and Thakor (2015) argues that banks asset portfolios are opaque from a shareholder perspective, and they also find empirical support using bank balance sheet observables for their predictions. A growing body of work focuses on the consequences that arise from opacity. Jones et al. (2012) examine the implications of opacity for price discovery and financial stability. They show that opacity plays a crucial role for contagion that has potential to destabilize the financial system. Subsequent work by the same authors further reinforces this view by showing that banks investments in opaque assets increase systematic risk and reduce idiosyncratic risk which ultimately creates price synchronicity and systemic risk (Jones et al. (2013)). Jiang et al. (2016) ask whether the geographical deregulation of banking markets affected the quality of information disclosure to the public. They conclude that increased competition from the removal of entry barriers reduced earnings opacity which created better conditions for market participants to monitor bank behavior. The question of whether earnings opacity facilitates regulatory forbearance is explored by Gallemore (2013). He shows that forbearance correlates positively with earnings opacity. In contrast to these studies, our research analyses whether a change in debt priority structure which shifts monitoring incentives away from depositors towards general creditors that are considered to have more sophisticated monitoring technology can help reduce earnings opacity. 1 Given the adverse consequence of opacity reported in previous studies and our own results of a preliminary inspection of the data that documents negative effects arising from earnings opacity for bank soundness and profitability, we argue that the change in debt priority structure and the hypothesized increase in 1 General creditors are, most importantly, suppliers of Fed funds, trade creditors, beneficiaries of guarantees, holders of bankers acceptances, unsecured lenders, landlords, and counterparties to swaps and other contingent liabilities. 3

4 monitoring incentives for non-depositors can complement the monitoring effort by supervisory agencies to reduce earnings opacity. Second, our study helps inform the debate in the policy community. In fact, this research sheds light into the question of whether policy initiatives that reallocate monitoring incentives to investors with sophisticated information processing capabilities such as the introduction of convertible subordinated debt requirements and the depositor preference laws in Europe in 2014 will work in practice. Third, understanding whether changes in debt priority structure limit the scope for discretionary accounting choices is also important from a macroeconomic perspective. Laeven and Majnoni (2003) show that rules for loan loss provisions models based on incurred losses can exacerbate pro-cyclical lending behavior. Building on this insight, Beatty and Liao (2011) illustrate that this problem is particularly acute for banks that delay recognition of expected losses because banks with more timely loss recognition tend to reduce lending less during recessionary periods. Fourth, our experiment also matters quantitatively. Banks in the 15 states that enter our main analysis account for 53% of total banking system assets in the U.S. State-chartered banks hold 47% of total banking system assets in these states. The share of uninsured deposits in these banks is 9% of total liabilities, and non-deposits account for more than 14% of total liabilities. That is, more than 23% of the liability side of a banks balance sheet is affected by these laws. Our key findings are based on widely used earnings opacity measures used both in the accounting literature (e.g., Beatty and Liao (2014)), and in the finance literature (e.g., Jiang et al. (2016)), where higher absolute values of the residuals from a regression of LLP on a set of independent variables are interpreted as more intensive earnings management, and, consequently, greater earnings opacity. We find robust support for the view that the subordination of general creditors claims below depositors claims reduces earnings opacity. This finding obtains irrespective of the way we measure earnings opacity, including the use of earnings opacity measures that focus on other real estate owned relative to total loans (OREO), banks opaque assets, and identifiers that provide information on missing data about loan loss provisions. The magnitude of the effect is also sizeable, ranging from 7.35% to 9.84% in terms of the reduction in earnings opacity. We present tests that explore the mechanism behind these results. A plausible explanation is that market discipline plays a key role for our findings because subordinating claims held by nondepositors increases their monitoring incentives. To examine whether this explanation holds true in our data, we follow the approach in the market discipline literature pioneered by Flannery and Sorescu (1996) and compare the pricing response of deposit and non-deposit funding to the change in debt priority claims. Consistent with the change in debt priority structure and the reallocation of monitoring incentives towards non-depositors whose claims are made riskier, we document that the cost of non-deposits increase by approximately 23%, and the coefficient estimate is highly significant. Our finding remains consistent with non-depositors imposing market discipline by monitoring banks more intensively. Moreover, if monitoring by junior claimants drives reductions in earnings opacity, these effects are likely to be more pronounced for the weakest banks in our sample. Again, this is what we find when we examine a subsample of banks close to insolvency, where we show that the coefficients increase in magnitude. Following this line of reasoning, we anticipate that those banks that are increasingly in the spotlight also restrict their discretionary accounting choices, and engage in less earnings management. Indeed, the reduction in earnings opacity following the change in debt priority is larger among banks with low Z-scores (ln), high leverage ratios, and low levels of 4

5 profitability. Taken at face value, these findings can be interpreted to be consistent with an increase in monitoring by junior claimants. One further concern remains. An alternative explanation for our results that we need to rule out prior to concluding that our findings are the result of tighter monitoring by non-depositors is that banks offer higher rates to non-depositors to deter these creditors from monitoring. That is, the increase in rates paid to non-depositors is compensation to discourage them from forcing changes in conduct. If this alternative hypothesis is true, banks with disproportionally high non-deposit interest expenses are unlikely to restrict their discretionary accounting choices more than those with, on average, low non-deposit interest expenses because non-depositors of these banks are already being compensated for the additional risk. We shut down this alternative channel and show tests that are inconsistent with this view because such bank behavior would be at odds with profit-maximizing behavior, and, moreover, we observe changes in bank conduct that would not arise if non-depositors would just be compensated for bearing greater risk. On the contrary, these tests indicate that banks with non-deposit expenses above the median reduce earnings opacity more than those with low nondeposit expenses. Second, we rule out confounding factors. Note that any omitted variable that could bias our estimates of the average treatment effect needs to coincide with the law changes and, importantly, differentially affect the state and nationally-chartered banks. Our setting contains 15 separate enactments of depositor preference laws, which makes it less likely that a coinciding factor biases the results compared to a typical difference-in-difference setup where there is only one treatment. We show a series of further tests that address possible omitted variable problems. One of them relates to the endogenous choice of the bank charter. We show that charter switches do not affect our inferences. Another concern relates to the fact that the sample period is characterized by episodes of banking turmoil with regional banking crises in the New England states, Texas, and the Savings & Loans (S&L) crisis also coincided with the adoption of depositor preference laws. Removing these states and testing for the effect of the S&L crisis leaves our inferences unchanged. Likewise, we demonstrate that the subsequent regulatory responses to these crises, i.e., the Financial Institutions Reform, Recovery, and Enforcement Act (FIRREA) in 1989, and the Federal Deposit Insurance Corporation Improvement Act (FDICIA) in 1991 play no role for our conclusions. A further concern arises from the deregulation of banking markets because Jiang et al. (2016) find that deregulation reduces earnings opacity. We test for whether there are differential effects for state and nationallychartered banks and rule out that this is the case. A final set of robustness tests rules out that simultaneity bias, anticipation effects, and differences in bank size between state and nationally chartered banks affect our results. The remainder of the paper is organized as follows: Section 2 explains the legislative history of depositor preference. In Section 3, we present summary statistics, and describe the identification strategy. Section 4 reports our main results. Section 5 offers robustness tests, and we also rule out a variety of alternative explanations, and Section 6 concludes. 2. Institutional Details 2.1 Bank liquidation guidelines and depositor preference laws Bank resolution in the U.S. is based on detailed guidelines laid down in the Banking Act of 1935 which assigns the Federal Deposit Insurance Corporation (FDIC) the role of the receiver which pays 5

6 off claimants in line with the guidelines of the Banking Act. Once a bank enters liquidation, its assets are transferred to the FDIC which identifies and satisfies creditors, with the objective to maximize the net present value for the creditors. In the absence of depositor preference laws, the FDIC, in its role as a receiver, is first in line to obtain part of the proceedings in the liquidation as compensation for its administrative expenses. Next in line are the claimants who hold collateralized claims. Depositors with account balances below the deposit insurance coverage limit follow next. Once these claimants are satisfied, uninsured depositors whose account balances exceed the insurance coverage limit and non-depositors are paid off. 2 That is, claims by uninsured depositors and non-depositors are equal in rank. At the bottom of the claim structure are subordinated debt holders. The remaining assets are distributed to shareholders. Several states amended the claim structure for banks that operate with a state charter by changing the debt priority structure in their respective state banking laws. In 1909, Nebraska was the first state that assigned uninsured depositors priority claims over the claims of general, i.e. non-deposit, creditors. Subsequently, 29 other states also mandated depositor preference laws by elevating all claims of depositors, irrespective of their insurance status. Of those 29 states, 15 states (Arizona, California, Colorado, Connecticut, Florida, Hawaii, Kansas, Louisiana, Maine, Minnesota, Missouri, New Hampshire, North Dakota, Rhode Island, Texas) legislated depositor preference during our sample period 1983Q1 1993Q2. Table 1, Panel A, provides an overview about the staggered nature of depositor preference legislation. [TABLE 1: Year of adoption of depositor preference laws] To understand how these legal provisions are embedded in state laws, and gain an overview about the motivation behind the adoption of these laws, we screen the text of the law in all 15 U.S. states, and contact the legislative council archives and the chartering authorities. 3 We find that the provisions concerning depositor preferences are presented under headings entitled Involuntary Liquidation Procedure, Payment of Claims, or Distribution of Assets. Although the exposition differs somewhat across states, the priority structure for the claims on failed state-chartered banks that emerges looks as follows: 1. administrative expenses of the receiver; 2. secured claims; 3. deposits, both insured and uninsured; 4. general creditor claims; 5. subordinated creditor claims; 6. shareholders. One may question why not all states decided to assign uninsured depositors a preferential claim in case their bank is liquidated. To explore this, we plot in Figure 1 the states that adopted these legal guidelines to see if there is any geographic clustering. In Figure 1a, we show which states introduced these guidelines prior to 1993, they are shaded in dark grey, and Figure 1b highlights in dark grey those states that we include in our empirical tests. The diagram does not indicate any geographic clustering of states that adopted depositor preference laws. Next, we use the information we obtain in assembly laws from the legislative councils archives, and read the legislative council s digests, the concurrencies of the state amendments, and we also use keyword searches in Lexis/Nexis, Factiva, 2 3 Non-depositors are trade creditors, beneficiaries of guarantees, holders of bankers acceptances, unsecured lenders, landlords, suppliers of Fed funds, and counterparties to swaps and other contingent liabilities. Utah and Virginia are omitted because they enacted depositor preference laws in 1983 but the absence of some variables and the annual sampling frequency in Call Report data prior to 1983 render inclusion of these two states in our analyses impossible. 6

7 American Banker, and the Journal State Legislatures, focusing on the 6 months prior to the day of the introduction of depositor preference. 4 [FIGURE 1: State depositor preference laws] Our survey of the documents from the legislative councils suggests that the policy community did not pay much attention to the adoption of depositor preference. An example from California suggests that these guidelines are viewed to be useful to fix omissions in previous legislation. To illustrate, the concurrency of the state amendment in California mentions that the current law did not contain any priority for the payment of depositors and creditors and liquidation expenses, and instead required pro rata settlement. The new law addresses this lack of a priority order. Further, our review of the archives also reveals that assigning depositor priority is expected to speed up purchase and assumption transactions, and allows depositors to access their funds more quickly in case of bank failure. While the legislative council s digests stress in a few instances that the changes will allow a more swift reorganization and dissolution of state banks, in several states it is not possible to detect a reason being stated for the amendments in the priority structure. Except for some sources that report that the then FDIC Chairman, William L. Seidman, gave in 1986 testimony to the Senate Banking Committee and recommended federal depositor preference to reduce failure cost, our the keyword search of the media sources, limited again to one year centred on the announcement day of depositor preference, also does not suggest a great interest by the media in depositor preference. 5 Additional details about the political and economic environment during the time when depositor preference was introduced are shown in our Supplementary Online Appendix A. Since it is plausible to hypothesize that banking sector conditions, political economy considerations and the macroeconomy may be important for changes in state banking laws, we run a further survey using keyword searches in American Banker, Lexis/Nexis and Factiva to identify coinciding events that happen during the time when depositor preference laws were adopted. 6 There is some evidence that depositor preference is more likely to be observed in states with Democratic Party governors. Likewise, depositor preference is more likely to be found in states where democrats control the upper and the lower house of the legislature. We find no evidence that banking sector conditions correlate in a systematic manner with depositor preference. Precisely, our media survey does not suggest that the adoption of depositor preference laws coincides with periods of either good or bad banking sector performance. For banks in Florida, Hawaii, and Minnesota, we find reports that highlight excellent business performance, and for Arizona and California we observe reports about bank mergers. Banking sector difficulties that arise from declines in real estate and energy prices are reported in the news for Texas and California, as well as for Maine, New Hampshire, Connecticut, and Rhode Island. 4 Our keywords are as follows: deposit obligation, depositor obligation, claims of depositors, claim structure, bank liquidation, depositor preference, priority of claims, priority claim, liquidation priority, liquidation regime, claims to be paid before those of general creditors, pari passu with general creditors, deposit rank, depositor rank. 5 The American Banker, 14 th March 1986, has a news item with the title FDIC offers plan to protect uninsured depositors. Similar news reports are published in the same week in the Chicago Sun Times, the New York Times, and on Dow Jones Newswire. 6 The keywords are: bank OR banking OR financial institution OR depository OR budget deficit OR budget surplus OR oil shock OR unemployment benefit OR S&L crisis OR Savings and Loans crisis OR Federal Savings and Loan Insurance Corporation OR FIRREA OR Merger and Acquisitions OR M&A OR Dividends OR Earnings OR Deregulation OR Expedited Funds Availability Act. 7

8 2.2 Why do states adopt depositor preference laws? Econometric tests Our review of the evolution of depositor preference laws and the political and macroeconomic environment makes it appear unlikely that banking sector conditions motivate the introduction of these laws. Nevertheless, econometric rigour requires that we establish that the adoption of depositor preference was exogenous with respect to earnings opacity, and the other outcomes we are interested in. We therefore model the adoption of state depositor preference laws as a function of changes in variables over time and across states that capture private interest-group factors, public interests, and political factors. Using a linear probability model we estimate the following reduced form 1 = , where is a dummy equal to 1 if state introduces depositor preference in quarter, 0 otherwise; describes private-interest variables such as the share of bank assets held by state-chartered banks within the state-quarter; is a proxy for the intensity of the S&L crisis (captured by the ratio of assets in failed thrifts to total bank assets), total assets in failed banks (ln), and the mean bank profitability in the state; is a dummy that is equal to 1 if a democrat is state governor, 0 otherwise, to reflect differences in political parties willingness to regulate the financial sector (Kroszner and Strahan (1999)). State and quarter fixed effects are represented by and, respectively; is the error term. Our data spans the period 1983Q1 to 1993Q2 for the 15 states that introduce depositor preference, in line with the subsequent tests. Standard errors are clustered at the state level. We present this test in Panel B of Table 1 where we refute the view that private, public, or political factors drive depositor preference, and this inference obtains irrespective of whether each interest vector enters individually or simultaneously. Together with the evidence from our survey of the archives and media sources, these tests suggest that that the introduction of depositor preference appears to be unrelated to banking sector conditions or other concerns regarding the transparency of bank balance sheets. Having established the first one of the key assumptions necessary to deploy difference-in-differences estimation that the assignment of depositor preference appears to be as good as random, we turn to the second assumption: that the treatment and control group display parallel trends. In particular, this assumption requires that earnings opacity evolves similarly for treatment and control groups absent depositor preference. To test this assumption, we follow Whited and Roberts (2012) and use t-tests to inspect whether differences exist in the growth rate of earnings opacity between the treatment and control groups during each of the four quarters preceding enactment of depositor preference. These tests show that our control group is statistically indistinguishable from the treatment group. We cannot reject the null of equality in any instance. The assumption of parallel trends holds. Our control group constitutes a valid counterfactual. [TABLE 2: Parallel trends] 3. Data Description, Variable Definitions, and Identification Strategy We construct our quarterly bank-level data set using information for commercial and savings banks obtained from the Quarterly Report on Condition and Income (Call Report), provided by the Federal 8

9 Reserve Bank of Chicago. The sample covers the period 1983Q1-1993Q2. We chose this timeframe because banks were not required to report Call Reports on a quarterly basis prior to 1983 and because all banks, irrespective of their charter, were subject to depositor preference laws from 1993Q3 onwards as a result of the Omnibus Budget Reconciliation Act. Following Osterberg (1996), we exclude banks operating in New York state due to their unique size and regulatory environment. We ensure that only institutions that operate in at least four quarters prior to and after the introduction of state depositor preference laws are included so that a sufficiently large number of observations for each bank enter our empirical tests. Applying these sample screens, we obtain a final sample consisting of 159,156 bank-year observations for 6,906 banks. 7 The Call Reports provide information for each bank on charter status (state or national), location, size in terms of total assets (ln), the ratio of equity capital to total assets (ln) to measure capitalization, total loans, loan loss provisions, the ratio of non-performing loans to total loans, the value of other real estate owned by the bank, return on assets (ln), and return on equity (ln). We calculate deposit interest expenses as the ratio of deposit interest expenses to total deposits. Similarly, we compute non-deposit interest expenses as the ratio of non-deposit interest expenses scaled by total non-deposit liabilities. We combine this information with data from the Case-Shiller Index and data on the statelevel per capita income growth rate and unemployment rate from the Federal Reserve Bank of St Louis and merge these variables into the bank-level data set. 3.1 Measuring bank opacity To measure bank opacity, we compute a number of different proxies for earnings opacity. The literature on earnings opacity in banking is vast. We follow the best practice described by Beatty and Liao (2014) which is widely used in accounting and in recent finance literature by Jiang et al. (2016). All these measures of earnings opacity are based on the concept of discretionary loan loss provisions, calculated as the absolute values of the residuals from a regression of loan loss provisions on a set of independent variables. The motivation behind this idea is that discretionary loan loss provisions are one of the most commonly used methods through which banks manipulate both earnings and regulatory capital (Jiang et al. (2016)). Discretionary loan loss provisions therefore constitute indicators of the abnormal accrual of loan loss provisions. Higher values are interpreted as evidence of more intensive earnings management and greater earnings opacity ((Beatty and Liao (2014); Jiang et al. (2016)). Given that a variety of proxies for earnings opacity exist, we follow the preferred earnings opacity measures proposed by Beatty and Liao (2014). The central feature of the first model presented by Beatty and Liao (2014) is that it allows separating out the systematic component of loan loss provisions (driven by state-specific and bank-specific determinants) from the abnormal component, which proxies for discretionary changes in loan loss provisions. The larger the abnormal component, the higher earnings opacity: a large absolute value of the residuals therefore reflects changes (either positive or negative) in loan loss provisions that are related to a discretionary manipulation of loan loss provisions (Jiang et al. (2016)). We run the following regression (2),, = 1,, +1+ 2,, + 3,, 1+ 4,, 1+ 5,, + 6,+ 7 S, + 8, + 9ST + b,,, 7 Our findings remain unchanged when we include New York banks in the sample. Likewise, when we include banks that do not operate at least four quarters before and after the law change in the sample the results are virtually identical 9

10 where,, is loan loss provisions in bank b in state j in quarter t. d,, is the change in nonperforming assets. As in Bushman and Williams (2012) and Jiang et al. (2016) we consider the first lag and lead of d,, to capture both historical changes in NPA as well as current and forwardlooking information on NPA when banks choose the current level of LLP.,j, 1 is the log of total assets in the previous quarter, and reflects differences in the level of monitoring by regulators and private-sector stakeholders of banks of different size.,j,t is the change in total loans in quarter t divided by total loans in quarter t 1.,t (the Case-Shiller Index), S,t (the change in Gross State Product) and, (the change in the state unemployment rate) capture the effects of time varying state-specific elements of the macroeconomic environment that may affect LLP. Importantly, our tests include state fixed effects, ST j to capture unobservable time invariant statespecific characteristics. Our discretionary LLP variable is therefore LLP1 b,j,t =,,. Given the abundance of earnings opacity measures Beatty and Liao (2014) suggest using alternative modifications of Equation (2). Using the same approach, we consequently also construct LLP2 b,j,t and LLP3 b,j,t based on the following two equations, respectively: (3),, = 1,, +1+ 2,, + 3,, 1+ 4,, 2+ 5,, 1 + 6,,+ 7, + 8 S, + 9, +,,, and (4),, = 1,, +1+ 2,, + 3,, 1+ 4,, 2+ 5,, 1 + 6,,+ 7, + 8 S, + 9, + 10ALW,, 1 +,,t, where ALW,j, is the loan loss allowance over the quarter divided by total loans. Finally, we use the approach proposed by Bushman and Williams (2012) which relies on estimating the following reduced form (5),, = 1,, +1+ 2,, + 3,, 1+ 4,, 2+ 5,, S, +,,t, to construct LLP4 b,j,t. To ensure that our inferences are not driven by peculiarities of the loan loss provisioning approach, we also consider two additional measures of earnings opacity suggested elsewhere in the literature. First, we construct MISSINGb,j,t, a dummy variable equal to 1 if a bank reports a missing value for,, in the Call Report, 0 otherwise. Nier and Baumann (2006) highlight that missing values for loan loss provisions are a good measure of earnings opacity because they indicate that banks are actively choosing not to reveal potential losses. Second, we calculate the ratio of OREO to total loans. The intuition behind this measure is that it is difficult for outsiders to value real estate owned by a bank. The higher this ratio, the more opaque is the bank (Flannery et al. (2003)). Third, following Flannery et al. (2013) we calculate OPAQUE,, which is the sum of the book value of bank premises and fixed assets, investments in unconsolidated subsidiaries, intangible assets (such as mortgage 10

11 service right and core deposit intangibles) and the balance sheet category other assets (such as accounts receivable, repossessed autos, boats and other collateral). The intuition is that opacity is positively related to OPAQUE because investors find it difficult to value such assets. Finally, we create a small loss dummy variable which is equal to 1 if 0.1 <0, 0 otherwise. The intuition is that less opaque banks are more likely to report small losses rather than engage in earnings smoothing (Barth et al. (2006); Lang et al. (2003); Lang et al. (2006)). 8 Table 3, Panel A, provides an overview about the variables used in the analysis and Panel B of Table 3 tabulates the summary statistics. [TABLE 3: Summary Statistics] 3.2 Preliminary Inspection In light of the adverse consequences arising from opacity highlighted above, one would expect opacity to correlate with bank risk taking, the quality of the loan portfolio, and profitability. Prior to establishing the causal effects of the changes in debt priority structure, we offer a preliminary inspection of our data set based on correlations between earnings opacity and Z-scores (ln), the nonperforming loans ratio (NPL), return on assets (ROA, ln), and return on equity (ROE, ln). The correlations presented in Panel B of Table 4 support our expectations. In line with policy makers views, these simple tests indicate that measures which motivate banks to reduce earnings opacity may enhance profitability, limit risk taking, and engender greater bank stability. [TABLE 4: Correlation between earnings opacity and bank performance] 3.3 Identification Strategy To tease out the causal effects of changes in debt priority structure on earnings opacity, we employ difference-in-differences estimation. This approach compares the cross-time evolution of earnings opacity within treated banks (with state charters) and the group of control banks (with national charters) in the same state following the adoption of depositor preference laws. Accordingly, we estimate the equation (6) y,, = βb + β1charter, + β2dplj, *Charter, + β3xb,j, + γj, +ηb,j,t, where y b,j,t is a dependent variable for bank b in state j at time t, that is, one of the eight proxies for earnings opacity described above. DPLj,t is a dummy that is equal to 1 for all banks in states and quarters following introduction of depositor preference, 0 otherwise; Charter b,t is a dummy variable equal to 1 for all state-chartered banks, 0 for nationally-chartered institutions; X b,j,t is a vector of banktime varying control variables; βb and γj, are bank and state-quarter fixed effects, respectively. 9 These 8 Note that Jiang et al. (2016) also use earnings restatements triggered by the Securities and Exchange Commission (SEC) to measure opacity. While our sample period predates the release of information on earnings restatements by the SEC, and we therefore cannot offer such tests, the lack of information on restatements is unlikely to yield additional insights because Beatty and Liao (2014) show that the above earnings opacity measures are highly effective at predicting earning restatements. As discussed in the introduction, alternative measures of bank opacity based on ratings disagreements for bank bonds (Morgan (2002) and market microstructure properties of bank share prices (Flannery et al. (2004, 2013)) also exist in the literature but employing them in our analysis is beyond the scope of our research. 9 The dummy variable for depositor preference is captured by the state-quarter fixed effect and consequently does not appear directly in the estimation equation. 11

12 dummy variables account for unobservable time-invariant bank-specific and state-time-varying factors that are common to all banks. The error term is ηb,j,t. The coefficient of interest is β2, which measures the effect of depositor preference for statechartered banks. We cluster heteroscedasticity-adjusted standard errors at the bank level to account for serial correlation within each panel although, as we show below, the findings are invariant to alternative clustering procedures. 10 Bank fixed effects eliminate all time invariant bank-specific sources of unobserved heterogeneity such as location and managerial characteristics. The state-quarter fixed effects have two attractive properties. First, they eliminate all time varying state-specific confounding omitted variables that simultaneously affect the treatment and control groups such as changes in per capita income, unemployment rates and demand-side effects. Importantly, they also ensure that the average treatment effect (ATE) is estimated based on comparisons between the treatment and control group within the same state-quarter. In other words, our treatment effect represents the reactions of banks that operate in the same macroeconomic environment to depositor preference. 3.4 Exogeneity of Depositor Preference Laws [TABLE 5: Exogeneity Tests] Before conducting formal tests we run a number of diagnostic tests to rule out endogeneity concerns arising through simultaneity bias. Specifically, we investigate whether enactment of depositor preference law was to some extent related to bank opacity. Using a linear probability model we estimate the equation (7) DPL j, t = α j + β k, + Z j, t + π t + μ j, t, where DPL j, t is a dummy equal to 1 in the quarter state j enacts depositor preference law, 0 otherwise;, denotes earnings opacity defined by equations (2), (3), (4) and (5) above and is the average level of earnings opacity in state j during the quarter when depositor preference is enacted (k = 0) and the three preceding quarters ( 1,3 ); Z j, t is a vector of state-level macroeconomic and banking characteristics; α j and π t are state and quarter fixed effects, respectively; μ j, t is the error term. Observations from quarters after depositor preference laws are enacted are excluded from the estimation. Simultaneity bias implies that LLP motivate enactment of depositor preference and one would therefore expect the β k coefficients in Equation (7) to be significant. However, this is not the case in Table 5. Rather the coefficient on the LLP variables is insignificant in all columns of the table, irrespective of whether control variables are included in the model. The econometric evidence decisively refutes the possibility that endogeneity bias arises through this channel. 4. Main Results We proceed in two steps. First, we examine the effect of debt priority changes on earnings opacity. Second, we analyse the mechanism that underpins these results. 10 In most of the estimations the regressions restrict the sample to banks in the 15 states that enact depositor preference laws during the period from 1983Q1 to 1993Q2. However, the results are robust to including banks from all U.S. states in the estimations. 12

13 4.1 Debt Seniority and Bank Opacity Table 6 reports estimates based on the sample of 15 states. This allows clean identification of the average treatment effect within the state-quarter and provides strong support for the hypothesis that changes in debt priority structure lead to reductions in banks earnings opacity. Columns 1 to 4 rely on earnings opacity measures based on loan loss provisions as the dependent variable. Irrespective of these alternative ways of measuring earnings opacity, our key coefficient of interest, the interaction term between the state charter dummy and the dummy for depositor preference enters negatively and is highly statistically significant. The economic magnitude is similar across specifications, indicating a 7.35% to 9.84% reduction in earnings opacity. Importantly, these results are invariant to estimating different specifications of Equation (6) that use separate state and quarter fixed effects as shown in Table B.1 of our Supplementary Online Appendix. Likewise, a less clean econometric setup that includes banks located in all 50 states in an alternative and larger sample shown in Table B.2 of the Supplementary Online Appendix also leaves our inferences unchanged. [TABLE 6: Depositor Preference and Earnings Opacity] The control variables generally display consistent signs and levels of significance in all tests. Large banks tend to be significantly more opaque than smaller ones whereas earnings opacity is negatively associated with the capital ratio. Moreover, banks that incur losses report significantly higher levels of earnings opacity. Our estimates also indicate persistence in earnings opacity through time. The lagged dependent variable is positive and significant in Columns 1 to 4 in Table 6. Finally, the state charter dummy variable is insignificant throughout Table 6. State chartered banks do not engage in more or less earnings management relative to nationally-chartered institutions. 4.2 What drives the reduction in earnings opacity? Next, we explore the factors that underpin the observed reductions in earnings opacity. Recent theory by Birchler (2000) offers a plausible explanation. He argues that changes in debt priority reallocate monitoring incentives to those creditors whose claims are subordinated resulting in greater incentives to impose market discipline. In line with this argument, one would expect these creditors to demand higher risk premiums in return for the additional risk they bear. Since this demand for higher risk premiums increases funding costs, banks are likely to take steps to change their behavior and engage in less earnings management. Our tests that aim to establish whether market discipline lies at the heart of our findings follows the convention in the market discipline literature by comparing the interest rate response of non-deposit funding to the change in the debt claims riskiness (e.g., Flannery and Sorescu (1996)). 11 To compute these tests, we rely on Equation (6) with the sole modification that we use the ratio of non-deposit interest expenses to total non-deposit liabilities as the dependent variable. The results of this test are provided in Column 1 of Table 7. The coefficient on the interaction term between the state charter dummy and the depositor preference dummy is positive and significant at the 1% level. Consistent with their more junior status following enactment of depositor preference, we find that the cost of non-deposits increase by approximately 23%. This finding is our first piece of evidence that is 11 Call Reports do not report separate information on interest expenses on insured and uninsured deposits during the sample window. However, deposit interest expenses include both categories. We therefore use deposit interest expenses as a proxy for the cost of uninsured deposits. 13

14 consistent with the idea that non-depositors engage in monitoring and that this is the mechanism behind the reduction in banks earnings opacity. On the other hand, uninsured depositors claims are made senior which tends to have the opposite effects and dampen market discipline. The estimates in Column 2 of Table 7 show that conferring seniority upon uninsured depositors causes a significant reduction in deposit interest expenses that is equivalent to a 0.8% decrease. [TABLE 7: Funding costs] To dig deeper into the monitoring hypothesis, we analyse whether the extent of the increase in nondeposit funding costs correlates intuitively with cross-sectional characteristics of banks. Precisely, if monitoring drives our results, weak banks, i.e., those closer to insolvency, should be subject to greater scrutiny by debtholders, and such debtholders are likely to respond more strongly to changes in priority structure. In other words, the increase in non-deposit interest expenses should be greater in magnitude for weaker banks relative to healthier banks. Our empirical test for this hypothesis classifies banks whose Z-score (ln) is equal to or below the median, and banks whose leverage ratio is above the median as weak banks. All remaining banks are considered to be healthy. We then replicate the difference-in-differences regressions with non-deposit interest expenses as the dependent variable. The remainder of Table 7 presents the results. Consistent with this intuition, our analyses highlight the increase in non-deposit funding costs to the change in debt priority structure is more pronounced among weak banks. While the key coefficient is insignificant in Column 3 when the sample is restricted to banks operating with Z-scores (ln) above the median, for banks with Z-scores (ln) less than or equal to the median in Column 4, the interaction coefficient enters significantly, signalling a 67% increase in non-deposit funding costs. These patterns are mirrored, although not so pronounced, in the subsequent columns where we define weak banks based on leverage. State-chartered banks with leverage values above the median experience a significant 34% increase in non-deposit funding costs after depositor preference is enacted. We present the corresponding tests for earnings opacity in Table 8. One would expect that banks which are subject to the greatest increase in monitoring take larger steps to reduce earnings opacity. Indeed, this is precisely what we find. Throughout the table, the economic magnitude of the average treatment effect is larger for weaker banks relative to healthier banks. The reduction in earnings opacity following the change in debt priority is larger for banks with low Z-scores (ln) and high leverage. When we broaden the definition of bank health and examine profitability in terms of ROA and ROE in Column 5 8, we uncover larger reductions in earnings opacity among less profitable banks. While these findings are consistent with the idea that monitoring may be responsible for the patterns in our data, we have yet to rule out alternative explanations. [TABLE 8: Further Evidence for Earnings Opacity] A way to test whether banks engage in earnings smoothing is to investigate under-reporting of small losses in favor of small earnings (Leuz et al. (2003), Degeorge et al. (1999), Burgstahler and Dichev (1997)). This happens because managers can turn a small loss into a small profit thanks to a certain degree of reporting discretion allowed by accounting standards. This phenomenon, called 14

15 small loss avoidance, may lead to delayed recognition of small losses, and a higher frequency of small earnings. In Figure 2 we report the distribution of the ROA for state- and national-charted banks before and after DPL. For state-chartered banks, small losses become more likely after DPL, while small earnings become less likely. For nationally-charted banks, however, this is not the case, suggesting that DPL reduces the degree of small loss avoidance. [Figure 2: Distribution of Small losses] 4.3 Alternative Explanations Prior to concluding that monitoring is the driving force behind the reduction in earnings opacity, we need to shut down an alternative channel that may also be consistent with the increase in noninterest expenses. As stated above, our tests for monitoring are common in the market discipline literature which interprets correlations between higher interest expenses and greater bank risk as evidence of monitoring (e.g., Flannery and Sorescu (1996); Martinez Peria and Schmukler (2001); Goldberg and Hudgins (2002)). However, banks may simply pay higher rates to non-depositors to deter these creditors from monitoring, i.e., the increase in rates could be viewed as compensation for non-depositors to discourage them from scrutinizing earnings management too closely. If this alternative story were true, banks whose non-deposit interest expenses are greater than the median would be unlikely to respond to depositor preference laws and restrict their earnings management more than those with, on average, low non-deposit interest expenses because non-depositors of these banks are already being compensated for the additional risk. Table 9 presents tests that empirically refute this alternative explanation. Banks with non-deposit expenses above the median display greater sensitivity to the change in debt priority and reduce earnings opacity more than banks with below median values. [TABLE 9: Alternative Explanations] 5. Robustness Tests We offer a series of sensitivity checks to ensure the robustness of our findings. First, in light of the large literature on bank earnings opacity and given the myriad of measures proposed, we test the robustness of our results to using alternative variables for earning opacity. Second, we deal with other threats to identification by ruling out confounding factors. Perhaps the most serious identification threat in our setting are shocks that coincide with the enactment of depositor preference laws. Importantly, such factors are netted out by the state-quarter fixed effects. For an omitted variable to bias our estimates of the average treatment effect, such an omitted variable has to coincide with the law changes on the individual state level, and, additionally, differentially affect the treatment and control group. Since our setting contains 15 separate enactments of depositor preference law that serve as treatment, it is highly unlikely that a coinciding factor biases the results compared to a typical difference-in-difference setup where there is only one treatment. 5.1 Alternative Measures of Earnings Opacity Rather than using earnings opacity as the dependent variable, we test the robustness of the main results to using the ratio of other real estate owned to total loans, banks opaque assets and our 15